Research On Obstacle Detection Based On Binocular Stereo Vision

Binocular stereo vision is an importmant branch of visual research.Binocular stereo vision captures image information by using binocular cameras by imitating human cognition,and restores the scene's three-dimensional information according to the disparity principle.In recent years,with the rapid development of computer vision technology,binocular stereovision has been widely used in the fields of intelligent transportation,industrial detection,robot navigation,target recognition and cosmic exploration.This subject uses the indoor obstacle detection as the application background,collects the image information of the real scene in the room with the binocular camera as the object of processing,and performs the obstacle detection by matching the dense disparity information.This is of great significance for further research on robot positioning,path planning and intelligent navigation.The main contents of this research include camera calibration,stereo matching and obstacle detection.Accurate camera calibration is the basis for realizing stereo vision obstacle detection.Through the analysis of camera imaging principle,imaging model and commonly used camera calibration methods,combined with the research needs of the subject,Selected the calibration method of the binocular camera with Zhang's calibration method with low requirements for experimental equipment,simple operation,high precision,and good reliability,through experimental verification,using this method can obtain high precision camera calibration data.Stereo matching is the key to realize stereo vision obstacle detection.Through the analysis and introduction of the related theory of stereo matching and the commonly used stereo matching algorithm,combined with the requirements of obstacle detection on the accuracy of the disparity map,the stereo matching algorithm based on Bayesian theorem is studied in detail,the algorithm characterizes pixels based on grayscale gradients.Then this paper defines a new feature vector based on the color gradient,and uses this feature to improve the support point of the algorithm.Through experimental verification,the improved algorithm has obvious improvement effect on weak textures and disparity discontinuities.In the stereo matching process,the influence of mis-matching can be effectively reduced,and the accuracy of the generated disparity map is significantly improved.In this paper,an obstacle detection algorithm based on U/V disparity is introduced.The traditional UV disparity method can only detect the local information of obstacles,for this reason,the algorithm is optimized in this paper.The basic idea of optimization is that the disparity in different areas of the same object are very close,so iterative search can be performed within a certain range of disparity to gradually approach the true position of the obstacles.Then the disparity maps obtained by the original Bayesian algorithm and the improved Bayesian algorithm are used to perform obstacle detection experiments,and the experimental verification shows that the improved algorithm is superior to the original Bayesian stereo matching algorithm in the detection accuracy of obstacles.